canukstorm

Mike Wuerthele said:

22july2013 said:

Next: trifold phones.

I feel like I saw one at MWC.

danox said:

Who knows?

https://www.fortuneindia.com/technology/exclusive-perplexity-shuts-down-apple-acquisition-chatter-says-no-talks-underway/124260

And what is it? Other than the fact someone (a trusty venture capitalist) says they are worth 14 billion dollars are they a fancy web search engine? Once again if bought by Apple who’s staying?

https://en.wikipedia.org/wiki/Perplexity_AI Wikipedia says that they are a fancy web search engine which uses a variety of AI models (five) of them in fact if you’re running behind, does that sound like an easy road to balance five models at once, and get what you need done within a year or two?

danox said:

Highly unlikely 2025 is almost over Perplexity even if you finalized the deal in October/November, after you buy them, you’ll have to have many someone’s go through every single thing that they’ve done up until this point in even greater detail and to top it off most of the people at Perplexity are going to quit so what are you getting? That’ll take a good six months maybe even a year and it won’t help you at least not soon enough in fact, they won’t even help you at all, but you would’ve spent $14 billion dollars down the drain, but you can tell Wall Street that you’re doing something.

Sam at OpenAI did not get a check from Apple for a reason because there are no shortcuts in AI just like there are no shortcuts in designing operating systems and designing cpu/gpu chips particularly if you choose unwisely. Apples solution/path is on the edge. Most of their competition is working on a solution that phones home.


https://www.macstories.net/stories/hands-on-how-apples-new-speech-apis-outpace-whisper-for-lightning-fast-transcription/

https://garymarcus.substack.com/p/a-knockout-blow-for-llms  Eddie Cueless better talk to his researchers/programers on staff Sam, Google, Microsoft and Meta aren’t reliable.

https://garymarcus.substack.com/p/llms-dont-do-formal-reasoning-and

https://garymarcus.substack.com/p/a-knockout-blow-for-llms/comments

mpantone said:

It's clear that Apple senior management has come to the forlorn conclusion that the more ambitious use cases for Apple Intelligence aren't reliable enough for everyday use by Joe Consumer yet. They vaguely mentioned it "coming in 2026" but that could easily be Q4 2026 in iOS 27.2.

There's a growing consensus that LLMs (from all producers) don't have any reasoning capabilities. It's becoming more evident where LLMs fall short and have more limited usage cases that originally hoped for.

Apple can't release an LLM-powered Siri AI chat assistant that's only as good as ChatGPT or Claude. There's no differentiation no value add. The main benefit from Apple in that case would be privacy but Joe Consumer wants something that works better. So Apple needs to deliver both of functionality AND privacy. And building the AI model for that is hard especially if Apple is ethically sourcing data (i.e., licensing/paying content creators rather than stealing it).

There is also more talk these days of "model collapse." My guess is that Apple is very wary of this problem. We're already seeing some evidence that today's newer consumer-facing LLM models aren't really much better that those from just 12-18 months ago in overall performance except for a handful of edge cases (like programming).

In the long run, it's better for Apple to be cautious about its AI approach and prioritize reliability, accuracy and privacy over being Headline Of The Week. I really don't need another AI chatbot assistant that's wrong 40% of the time or one that suggests eating rocks or using glue as a pizza topping. That's just a wasted effort because it forces the end user to double- or triple-check everything (not just the blatantly wrong answers).

Consumer facing AI still has a handful of useful features. Some of the basic spell/grammar checking and basic message composition seems to be adequate. It's good at dealing with math and physics homework (it won't make you smarter at either subject). On iOS 18, Image Cleanup does a passable job as an integrated, on-device service. Yes, there are third party tools that do this. Yes, a Photoshop wizard can do this. But I'm not sure how trustworthy third-party tools are (in terms of privacy) and like many, I'm no Photoshop wizard.

Anyhow, I expect little new developments until WWDC 2026. I think what we saw in yesterday's keynote is all we're going to see for another 51 weeks. There will be new hardware before then but it's likely that they won't have any stunning new functionality.

At this point it's more important to me that Apple protects user privacy and security over fancy new features. I'll wait for those when they are fully baked with privacy and security at the forefront.

Disclaimer: I plan on installing iOS 26, iPadOS 26, Mac OS 26 around 9-10 months after they debut so not until June 2026 at the earliest. This is my usual M.O. I'm still on iOS 17, iPadOS 17, and Mac OS Sonoma. At this point, there's a chance I'm going to skip iOS 18 and Mac OS Sequoia (I skipped Crapalina).